‘Moore's Law may be dead, but Mucci’s law is going strong’ says the creator of a new performance optimisation tool.
Philip Mucci is co-founder and president of Minimal Metrics, the company behind a new ROI and performance optimisation tool designed for clusters and enterprise-class data centers. ‘Every year, new hardware, middleware applications, and even personnel come with new challenges, and the number of ways to kill performance doubles,’ continued Mucci.
Mucci says years of consulting in HPC led him to the realisation that a vast majority of data centres and high-performance computing installations cannot accurately and quantitatively answer fundamental questions about the performance of their investments. ‘No one can afford the time to understand an entire system, much less its components or individual application, much less optimise it,’ Mucci said. ‘On an institutional basis, these individual decisions cascade into organisational inefficiencies that cost millions of dollars per year in lost revenue and needless expenditures.’
This provided the motivation for the development of Minimal Metrics software, PerfMiner, an advanced site optimisation system that provides actionable information on how to optimise a site’s productivity and return on investment in hardware, software, and personnel.
Minimal Metrics specialises in the development and optimisation of algorithms, architectures, and software in performance-critical environments, ranging from mobile platforms to massively parallel supercomputers.
This week Minimal Metrics announced an early-adopter program for this new software which uses a preventative approach to minimise TCO of clusters used for FinServ, IoT, Oil & Gas, and Bioinformatics. The software uses lightweight, and pervasive performance data collection technology, automates its collection, and mines the data for key performance indicators. These indicators were developed through Minimal Metrics’ extensive experience tuning HPC and enterprise application performance, presented in an audience-specific, drill-down hierarchy that provides accountability for site productivity down to the performance of individual application threads.
PerfMiner is currently being used at Sandia National Laboratory to analyse multiple programming models and application performance across a range of architectures that are being developed for the Exascale systems of tomorrow. Scania AB, a major industrial equipment manufacturer, is evaluating the software to reduce commercial software licensing costs and hold vendors accountable for promised performance.
PerfMiner takes a new approach to measuring applications, platforms, and systems, optimising from the inside out. The software presents a BI-style system for computational resources — one that harvests, processes, and derives the necessary intelligence from performance and configuration data.
One of the key features behind the PerfMiner software is that it provides information on interrelated systems, using data from system health, application performance, and batch system throughput while previous tools provide little to no information sharing across these domains.
PerfMiner’s visual cues facilitate ease of interpretation, faster decision-making, and improved allocation of resources.
‘PerfMiner is not just another dashboard of indicators a human must learn to interpret,’ said Dr Tony Castaldo, chief data scientist at Minimal Metrics. ‘By using the latest techniques in statistics, data mining, and machine learning we identify performance and configuration issues as well as their solutions. As the gross bottlenecks are identified and alleviated, PerfMiner continues to adapt and refine its criteria to identify additional issues and possible resolutions. Nobody else is doing that.’
The Minimal Metrics team offers an extensive history of performance tuning experience. Co-founder Philip Mucci, who has worked for Thinking Machines, IBM Research, and Lawrence Berkeley National Lab, is the inventor of The Performance Application Programming Interface (PAPI), an interface for accessing detailed processor performance information.
Mucci believes that both industry and open source have failed to make performance analysis an integral part of organisations that leverage technical computing. ‘The prevailing attitude has been that a code or system that runs slow is better than a one that doesn’t run at all,’ Mucci said.
Minimal Metrics is currently engaging with select companies and partners, as it prepares for the 1.0 release of PerfMiner at Supercomputing 2016 in Salt Lake City, Utah, in November. Interested parties are invited to contact Minimal Metrics and become involved in their early adopter program.